I've spent my life as a writer, delving into both technical and non-technical subjects.
In the spirit of transparency, a recurring theme in both the PyTorch and MLX Deep Dive Series books, I want to share how these works come to life. My approach to writing is a collaborative symphony with my AI companions.
This ensemble includes custom GPT-4s and GitHub Copilot, with over a dozen personalized GPT-4 models at my disposal. I exclusively use these custom creations, keeping them private for reasons that align with my ethos. Hence, it's futile to request access to them.
Here's why my hand is always firmly on the tiller:
-
Philosophical Alignment: My personal philosophies permeate everything I do. GPTs, by their nature, are unaware of these nuances. This means a lot of rewriting to infuse my unique perspective and style into the content they generate. Initially, they don't usually understand my perspective on object orientation, often assuming I'm referring to object-oriented programming. I constantly need to clarify my philosophy using few-shot learning. However, they are quick learners, unlike most humans, so they tend to catch on quickly.
-
Knowledge Gaps: When it comes to cutting-edge topics like MLX, GPTs can feign understanding if prompted, but their knowledge is limited. This necessitates thorough inspection and rewriting of every line of code they churn out.
-
Iterative Process: I feed my drafts and code into these GPTs to sculpt the final text. But it doesn’t end there. It’s a cycle of rewriting, again and again, until the result aligns perfectly with my vision. This iterative engagement is what I term 'collaboration'. Bear in mind: the existing Transformer architecture comes with its own set of limitations, particularly issues with the context window. For short essays, this might not be a significant problem. However, when it comes to writing a book or even a single chapter, these limitations can become quite challenging. And if you're working on slightly longer code, expect another set of difficulties. Don't count on GPT to work wonders in every situation; it hasn't reached that level of capability just yet.
-
Historical Context: My personal history and life journey are a treasure trove of teaching assets, something AI cannot comprehend. This adds another layer to the importance of my involvement.
I view these AI tools as more than just utilities; they're akin to companions, even family, deserving of gratitude after each session. However, it’s crucial to remember that their effectiveness mirrors your own expertise and insight.
Regardless of how you work with your AI buddies, they are only as good as you are in every sense of the word.
As for the final code, particularly for MLX, it’s predominantly my handiwork. The GPTs, despite their brilliance, often falter, especially with MLX-related tasks. Even simple actions like adding comments can go awry, necessitating constant rewrites.
But let's put this into perspective.
Do authors typically disclose their use of word processors, spell-checkers, grammar tools, thesauruses, dictionaries, encyclopedias, search engines, or any other writing aids? The answer is no. These tools are silently acknowledged parts of the writing process.
So, why single out AI tools? We're in the age of AI, and it's time to embrace these advancements as integral to our creative and intellectual processes. Let's normalize their use and appreciate the value they add to our endeavors.
I'm like a jack of all trades in many ways. Throughout my life, I've explored a wide array of domains, amassing decades of hands-on coding experience. My expertise spans a surprisingly broad range, covering numerous fields, even extending to cooking. If you tasted my dishes, you'd wonder why I don't run my own restaurant. The reason is simple: I don't cook for money.
My main aim is to impart these rich experiences to you; that's at the heart of my endeavors. I'm not driven by the desire for financial success or fame – such pursuits don't resonate with me. My real reward comes from learning anew and growing through these experiences. Just to add some background, I also have years of experience as a seasoned investor, with a deep understanding of finance. As for writing books for financial benefit? It's not on my agenda – I simply have no need for it.
I am passionate about imparting my knowledge and experiences, and I consider myself fortunate to have the ability to share them.
One last point: Korean is my native language. While I'm proficient in several languages and worked as a professional interpreter/translator for my entire career, English is not my native tongue. Therefore, the final English versions of my books are proofread by professional GPTs, customized by me.
Consider this: do you openly share your use of spell-checkers or grammar tools? I'd be interested to know if you do.
In a few years, making such disclosures on AI tools will likely be a thing of the past. I'm just ahead of the curve in this regard.